Possibly a placebo issue, but could be interesting. And I have absolutely no experience in that matter :P
rwatch *(float *)0x5DA854
rwatch *(short *)0x5DA854
rwatch *(long long *)0x5DA854
To the GDB client, these are three distinct and valid watchpoints, which... when committed to the stub's runtime... become just one! They all compete for the same address key in a single std::map. Only the first breakpoint entry is retained at runtime (as per std::map::insert's behaviour).
And in theory (you can slap me if that doesn't happen in practice), if GDB decides to only remove some but not all breakpoints, then there will still be no run-time watchpoints left. Lol.